"ITAGAKI Takahiro" <itagaki.takahiro@oss.ntt.co.jp> writes:
> "Guillaume Smet" <guillaume.smet@gmail.com> wrote:
>
>> It's better but still slower than 8.2.
>
> It probablly comes from 'var-varlena' feature in 8.3. Now we store
> text fields in a compact format on disks and extract them on access.
> It consumes some CPU cycles. If all of data are in buffer cache
> and the encoding of database is single-byte encodings, the performance
> of LIKE in 8.3 was 10-20% slower than 8.2 on my tests.
Hm, it does seem I missed like.c when I converted all the text operators to
avoid detoasting packed varlenas. I'll send a patch in a few minutes to do
that. I'm surprised it would have such a large effect though.
-- Gregory Stark EnterpriseDB http://www.enterprisedb.com